-
-
Root folder
-
Degree1 burn
-
Degree2 burn
-
Degree3 burn
-
Healthy skin
-
FastAI
-
Basic CNN implementation
-
Training/Validation accuracy and loss
-
Xception transfer model
-
Predictions with Confidence level
-
Web interface
-
Prediction 1
-
Prediction 2
Inspiration
Accurate assessment of burns is increasingly sought due to diagnostic challenges faced with traditional visual assessment methods. While visual assessment is the most established means of evaluating burns globally, specialised dermatologists are not readily available in most locations and assessment is highly subjective. The use of other technical devices such as Laser Doppler Imaging is highly expensive while rate of occurrences is high in low- and middle-income countries. These necessitate the need for robust and cost-effective assessment techniques thereby acting as an affordable alternative to human expertise. Skin Burn is a vital problem which is often ignored, this is due to the unawareness regarding the detection of different degrees of skin burn. It is also one of the main cause of dying among unintentional deaths, which might be themselves the third leading purpose of demise globally. This has triggered the improvement of artificial clinical assistants, which might be used to song the physicians’ methods. Successful and proper remedy is primarily based on an correct analysis of damage depth and the motive of the burn. Burn location, intensity, and vicinity are the figuring out factors for the injury intensity.
What it does
Well, accurate assessment of burns is increasingly sought due to diagnostic challenges faced with traditional visual assessment methods. While visual assessment is the most established means of evaluating burns globally, specialized dermatologists are not readily available in most locations and assessment is highly subjective. The use of other technical devices such as Laser Doppler Imaging is highly expensive while rate of occurrences is high in low- and middle-income countries. These necessitate the need for robust and cost-effective assessment techniques thereby acting as an affordable alternative to human expertise. For the problem domain, I utilised a basic CNN model and various transfer learning techniques i.e- pre-trained model for classification of burn purpose. This methodology will also use a fine-tuning approach in which initial layers of these models will be frozen to extract useful features, and subsequently, top-most substituted layers will be trained using those features from the initial layers.
How I built it
Accurate assessment of burns is increasingly sought due to diagnostic challenges faced with traditional visual assessment methods. There are several existing pre-trained models that we will be using for transfer learning for example, XceptionNet, MobileNet, Densenet family, InceptionV3 etc. and machine learning technique like SVM for image classification purpose For the problem domain, we will be utilised a basic CNN model and various transfer learning techniques i.e- pre-trained model for classification purpose. Our methodology will also use a fine-tuning approach in which initial layers of these models will be frozen to extract useful features, and subsequently, top-most substituted layers will be trained using those features from the initial layers.
In the end we will build a simple web application using Flask, HTML, CSS and JS in which the user would choose a burn image and the model would give accurate prediction regarding the same.
Challenges I ran into
Dataset is small, about a 1k image samples (difficult to get data in the field of AI in Medical Imaging) Difficulty in getting patients confidential data fom various hospitals. Scraping of images for every body part Model is extremely inefficient Only works on burns located on the skin. Proper Data processing was to be preformed. Can use image segmentation/object detection to improve results .
Accomplishments that I'm proud of
This project can do have a major impact in the world of Artificial Intelligence in Medical Sector. Solved a problem which can solve a problem on a larger scale. Was able to increase the accuracy of the model to minimum loss value on the testing dataset.
What I learned
Improvements to models. Improvements to the dataset. Techniques used in data processing. Contributing something in the medical industry working in this sector. Discovering ways for meeting the needs of Dermatologists. How this methodology can be used in combact situations. for e.g: War which happend between Russia and Ukraine, many people would have suffered from burn injuries and a proper treatement at that particular time could have saved many lives, that is by using a tool(similar to Infrared Thermometer) which could detect the degree of the burn and cluster people on the basis of the burn, meaning a person suffering from a Degree1 burn can be treated minorly and one suffering from degree3 burn is to be rushed to the hospital.
What's next for Multiclass Burn wound Classification using Deep learning
- Mobile application: Develop a mobile application that can be used by medical practitioners to get instant results.
- AR application : A mobile application that uses camera and provide live results on the various degrees of burns and their location.
- Research Paper implementation.
- Business idea: Can be made proprietary and provide access to only hospitals/other medical institutes with proper doctors consultation.
- A complex web interface where the user can upload the images of the burnt body parts, the platform would results from the model.
- Bulding a DALI model which would differentiate between skin burn and skin rash and various other skin realted diseases.
Log in or sign up for Devpost to join the conversation.